Passive WiFi CSI Sensing Based Machine Learning Framework for COVID-Safe Occupancy Monitoring

Abstract

The COVID-19 pandemic requires social distancing to prevent transmission of the virus. Monitoring social distancing is difficult and expensive, especially in "travel corridors" such as elevators and commercial spaces. This paper describes a low-cost and non-intrusive method to monitor social distancing within a given space, using Channel State Information (CSI) from passive WiFi sensing. By exploiting the frequency selective behaviour of CSI with a cubic SVM classifier, we count the number of people in an elevator with an accuracy of 92%, and count the occupancy of an office to 97%. As opposed to using a multi-class counting approach, this paper aggregates CSI for the occupancies below and above a COVID-Safe limit. We show that this binary classification approach to the COVID safe decision problem has similar or better accuracy outcomes with much lower computational complexity, allowing for real-world implementation on IoT embedded devices. Robustness and scalability is demonstrated through experimental validation in practical scenarios with varying occupants, different environment settings and interference from other WiFi devices.